Ask HN: What Are You Working On? (July 2026) What are you working on? Any new ideas that you're thinking about? |
Ask HN: What Are You Working On? (July 2026) What are you working on? Any new ideas that you're thinking about? |
I'm also working on an update to ShopifySharp, the .NET package I maintain for Shopify's graphql and rest APIs. I need to regenerate the graphql types and the fluent query builders for the July 2026 API version that was just released, and I'm planning on some extra QoL improvements that I've run into while using the package over the last couple of months. I particularly want to add some F# QoL features, since I wrote the package in C# but use F# in all my personal projects. (https://github.com/nozzlegear/shopifysharp)
StudyEngine is a webapp I'm using while doing my masters in comp sci. I upload lecture notes, textbooks, papers, etc. It then extracts topics and tracks my mastery of them over time. It uses an LLM to generate questions and flash cards. It loops in some newer learning science ideas. It tests recognition first(multiple choice), and then once a level of mastery is matched, it switches to recall. Working on adding RAG to it, so I can surface where in the source material something can be reviewed when going over quiz results. Currently just for me an some friends. If can get a good eval set up, I might work on optimizing cost and seeing if it could be opened up.
NomNominees is simple webapp that tracks James Beard, Great American Beer Festival, Festival of Barrel Aged Beers, and other awards. I use it when I'm traveling to find places to check out. Even just a cluster on a map shows me neighborhoods I might want to check out.
The goal is to find on-chain structural anomalies, as well as seeing if clustering by behavior has emergent semantic properties
It helps me to automatically save a tab that's not been used in a while so it auto-closes it but saves it as well as having the ability to snooze a tab like how you'd do it in gmail.
Everything is locally stored with 100% privacy in mind.
And vim like navigation is natively done.
Underpinning my current app is an e2ee local-first sync engine, basically it is a traditional client-server sync (encrypted logs + snapshots sequenced with integers). It sends bundles of Loro CRDT operations. I wrapped the client side in WASM to power the web app and the CLI and have started a swift wrapper to port to native iOS. Bundle size is 3MB/1.2MB g-zipped so pretty happy with it. I've realised that web encryption is kind of bs (at least not as "WE CAN NEVER ACCESS YOUR DATA" as some vendors state) if someone else is distributing the app.
Over the last week I have done a lot of performance work & data remodeling - CRDTs are interesting because you can let data fall through the gaps if you're not careful.
This is an open source tool to run background coding agents + dev environment in isolated VMs. So far it has allowed me to migrate a majority of long running coding sessions to my homelab to run remotely. I can also run multiple in parallel without worrying about race conditions or my host machine breaking.
Any demos available of the web based browser?
Available at https://whynotlog.com and promo code HACKERNEWS gives access to the pro plan for six months.
I've found Astro to be an amazing framework for simple, performant websites. It stays really close to basic HTML and CSS while adding useful features such as scoped components, layouts, and easy Markdown blog integration.
So I have been using it to build websites. But many things keep repeating with every website I build, so I began working on this project to create a base that I can use for every new web project.
It references content from my Clean Web Development Guide: http://webdev.bryanhogan.com/
When it is far enough along, I will use it for the landing page of the app I'm working on: a customizable solution for self-tracking including habits, health and journaling, or whatever else you need: https://dailyselftrack.com/
After more than 400 days of traveling around Korea, Macau, Mainland China, Japan and Australia, I'm now returning to Germany / Europe looking for work. I wrote about that in my monthly mail-letter: https://bryanhogan.com/follow
As that DDoS was going on I realized that some of our dev and staging processes were impacted by it, and that apt-cacher-ng was doing nothing to help us.
apt-cacher-ultra snapshots the repo meta-data after verifying it, and only promotes it if the metadata all checks out. Additionally, it can optionally keep a list of "hot" packages, and can include those in the snapshot calculation.
Additionally, apt-cacher-ng would regularly choke and require some handholding. I'm hoping -ultra resolves that as well.
It is build using a model that can classify messages (ham/spam/marketing), packaged for Apple Mail but could be used in other places.
https://github.com/thegagne/aep-conformance-test
Did pretty well, only took a day or so. I first had it inventory every MUST, SHOULD, and MAY in the spec, and then let it rip. I did guide it quite a bit to get what I wanted, but at the end I’m pretty happy with it as a first draft.
Helped me learn the spec and will be helpful to hone my dotnet AEP server, and aepbase.
There already existed an aep e2e validator which does a similar thing, but this is more thorough and generates a nice report. It will tell you not just whether your API follows the spec, but also what parts of the spec it does not implement.
https://truetrials.substepgames.com
Previous comment: https://news.ycombinator.com/item?id=47749027
If you're an early stage b2b founder, I'd love to hear your feedback about TractionBeast.
[1] https://hyperclast.com/ - fast, self-organizing, self-hostable replacement for Notion
It’s a better code reader built on top of sem (treesitter). I’m getting a lot of massive PRs at work now, and this has helped a lot with reading them. It decomposes the changes into entities and sorts based on what has the most dependencies. This tends to put the most important functions first. Plus I can click through the dependencies for each function and mark things as reviewed as I’m reading them. It’s a big improvement over the GitHub review flow for me at least.
Everyone who plays D&D has experienced the moment where they forget key details about the collective story they’re building. From ‘hey it’s been a month, where are we?’ to ‘wait who was this crazy npc again?’, ai is excellent at transcribing, notetaking and building a knowledge graph of your fantasy world.
I’m still building mostly for myself by adding a ton of features I know my friends would want, but also think there’s some ‘there’ there.
The idea is simple: let Loracle record your sessions on discord or upload the raw audio of your sessions, then get a rich personal wiki and session notes you can interact with.
If you’re mid-campaign you can also upload session notes from plain text and it bootstraps a campaign wiki. Then future audio based sessions have a good base of npcs, quests, characters, etc to build off of.
At this stage I’d love feedback more than anything else. Happy to comp a lot of usage to HNers in return for some reports on how well it’s serving you. Email admin@loracle.app for anything and everything.
- Each component in a mini app in a heavily locked down container - Components are deployed and built in a web workspace, in the same workspace you can open a terminal and use your favourite coding agent to work on component code (each terminal is itself heavily sandboxes, has rw access only to the edited component code and users home dir) - Everything comes with heavy rbac and minimum permissions - Oh so much more
Explaining this well is hard, much like explaining to someone what Kubernetes or AWS does. This is at a level of what a sophisticated company infrastructure team would run, just as a workspace you can deploy for yourself easily and agents just build within that framework (I’m a cofounder of a infra/compute/datacenter startup and intimately familiar with this kind of complexity)
The main thesis is that Claw-style agents still feel like school projects, and that in the agentic era apps on demand will be more of a thing, and that the current systems weren’t built to deal with a whole new app built every few minutes.
May or may not end up as open source soon
It's connected to all papers of course, and all kinds of scientific simulators and specialised models. But I'm currently in Shanghai talking to labs to join a CloudLab (and hopefully setting up our own robotic labs), so that AI can actually order real physical experiments that are executed cheaply, efficiently and seamlessly as tool calls.
Through experiments like autoresearch we have seen that AI is already, if not always smarter, at least more systematic than humans at following the scientific method relentlessly (hypothesis-experiment loop). Let's see what we can do by connecting it to the real-world :)
I am building it on top of a new primitive called smolvm: a hybrid that combines isolation of VM with speed and flexibility of containers.
Most critical infrastructure orgs don’t have the budget to hire consultants, and even if they do, the deliverable is a deck, or a spreadsheet, or a PDF. We want to help any org of any size create a security regimen outside of these stale and disparate docs. For FREE.
Plus we have additional tools that we are building on top of the free software that will help in other areas besides policies and procedures. Like OSINT of any orgs operational and physical footprints.
If you're a parent to an autistic child(like me), I'd love to talk to you about this.
If you know anyone who has an autistic child, It would be super helpful if you could tell them about this game.
Thanks!
At its core, it uses quadtrees, and has affordances for arbitrary topologies. Check out the planet and donut-world demos!!!!
- https://hello-terrain.kenny.wtf/examples/torus - https://hello-terrain.kenny.wtf/examples/cube-sphere - https://hello-terrain.kenny.wtf/examples/raycast-character-c... (a little slow to load~)
I myself am the first medical doctor and full stack engineer in the history of my country (250 million), graduated as a doctor at age 25, and we have over 100+ users [all of which are medical/dental students and doctors], 10s of billions of seconds studying smarter, hundreds of millions of questions solved, and more.
Our Super App has subsystems including MedGPT, MedAgent, Spaci (our own take on spaced repetition) and much more.
We're bootstrapped, and continuing to scale. If you are in medical school or know someone who is, please reach out!
Version 2 is a significant upgrade, and is a bottom-to-top rewrite of both the backend server, and frontend app.
I’ve been using an LLM extensively, and it’s been a huge help. I have, however, also run into its limitations.
If you’re open to the idea of composing code blocks and ideas, plus some generative UI exploration, feel free to join!
Part of my job is to approve / reject MCP servers based on how secure they are and whether they are suitable for use in an enterprise environment. I was tired of my team being called the bottleneck to AI adoption, so I set out to automate the whole process.
I periodically collect the MCP servers and every new version from the Official MCP registry and assign them a score based on 29 distinct criteria like runtime guardrails (e.g. destructive tools, over broad permissions, rug pulls), SAST scans and transport & trust model.
As a result of this exercise, I found that 1 in every 10 MCP servers is pretty much unusable (score 40/100 or below). 18% of the popular MCP servers with 1000+ GitHub stars contain one or more security issues. 184 servers to date have changed their tool definitions after publication, which may indicate a "rug pull" attack.
I built this for security minded people who also want to be at the forefront of AI adoption and for security teams who are tired to be called the bottleneck.
Browsing the index is completely free, you only have to request an API key if you want automated, programmatic lookups for any workflow.
Feedback is always welcome!
Rally-X: https://linsomniac.github.io/rally-xy/
Tempest: https://linsomniac.github.io/teapot/
Dig-Dug: https://linsomniac.github.io/digger/
And not an arcade game, but a multi-player throwback to a multiplayer shooter game my team used to play called nSnipes: https://github.com/linsomniac/isnipes
iSnipes does require downloading and running a server, the others you just play on the web.
I follow a bunch of gaming rss feeds just to keep up with what’s new in the industry. Figured I’d take those and turn them into a news aggregator to put them all into one place. Threw in some game deals/affiliate to pay the web hosting bills (hasn’t paid for anything yet, lol).
It came from a frustration that I needed to switch between the browser and the IDE to navigate through the code and leaving comments on Gitlab at the company.
So I thought it could useful to create something and let it be accessible to the public as open source.
link: https://github.com/LuyandaLia/reviewflow
In a nutshell, it accepts draft comments, which can be modified and submitted.
It auto configs the env for Python as it uses FastAPI for calls to Gitlab.
It's my initial attempt. Suggestions, reviews, contributions are invited.
One love
I continue to grow my main product BoltAI[1]
[0]: https://inka.page
[1]: https://boltai.com
Sharable, real-time synced maps, Google Docs for maps basically.
I think the coolest part is the import feature where you can paste a link to a video or article and it pulls out places and enriches them with images and a description. You can also write your own notes, vote on places to go with friends, and apply colors. Right now I am working on user acquisition and experimenting with different marketing approaches.
OrcaBot was my Jan+Feb attempt to defeat the lethal trifecta whilst offering all the bells and whistles of a claw like sandbox: https://orcabot.com/blog#breaking-the-lethal-trifecta
This month I've been working on the free desktop version which is available as of today but probably carries a few too many bugs to not be worth promoting just yet.
https://github.com/keloran/tiny-dfr
Unfortunately due to the way GitHub defaults to creating prs in the parent fork, I have accidentally created a few invalid prs in asahi before I was ready, and now am banned from creating a good upstream one
Since the last update, I released everything that had been in testing since April, like gallery view, custom avatars, birthdays and, most importantly – autofill from link.
Now I'm preparing for a big launch – working on the landing page, SEO and onboarding experience. Here's what I've done so far:
1. I updated the landing page to actually tell users about the app and look presentable. I already see a big improvement in conversion
2. I added SEO crap to the landing page. This is painful for me, but sadly that's how Google Search works (it doesn't). It's paying off, too
3. I overhauled the onboarding experience, to make it smoother for new users
Two more features are still in testing; I plan to ship them before the release, but currently i'm not completely happy about them.
Was using this only for my self, but i think it might be interesting for other people as well.
I've also replaced Linear with a local sqlite-backed tool, added tooling to speed up code nav, and am building "no-slop", a tool for enforcing architectural guidelines on vibe-coded projects.
www.memoryplugin.com
I tend to print a lot of stuff to read while disconnected. This is a tool to help squeeze as much content onto a printed page as possible instead of printing 4 or more pages per sheet.
A good use of Claude slop I'd argue. Currently trying to figure out how to set up the site so that an LLM tasked with printing content through it can figure out how to use it in the best way.
Right now it copes with important open source libraries on the model of clang-format's configuration, which is a real trick given the partial elaboration you need (with backtracking). But that works.
mathlib4 is the final boss, I don't currently even have a plan without per-directory quirks files which is probably a nonstarter.
After some time I figured the best use of AI is to produce even more AI-related slop and spend my occasional 2 dollars on the deep seek model to do it.
Models are fun when given a stable identity and made aware of it.
It uses DuckDB to expose a sql query interface in the website itself because I wanted to give the freedom to just do something interesting with the data.
My friend John had an idea which I really liked so I added "john mode" which shows what he was suggesting :-D
I think that Hackernews might like it but honestly, I have probably just made it out for myself and also as something to just share casually with folks on hackernews and other websites and hopefully I am able to help people and myself in some way with this website.
Open to some feedback as usual (for mostly all my projects really) and thanks for reading and have a good day dear reader and hey perhaps give my website a try!
I know what motivates me: seeing progress. The feedback loop of "do X, see Y gain" is what keeps me going.
So I started building an integrated dashboard that can aggregate data from multiple systems:
- My digital scale
- Apple Watch (sleep + running performance)
- Beastmaker Motherboard, which is an electronic board that you attach a hangboard to and it shows you various stats like how much force you're applying
The idea is that every morning I'll open the dashboard and be able to see exactly how much progress I've made the previous day: weight loss, strength gain, cardio performance.
It's an interesting problem. There's essentially two parts to it: Apple Health, which aggregates data from the scale and the Apple Watch and can POST-export it hourly, and the electronic board, which sends data via BLE in real time. The destination for both of these will probably be an always-on Raspberry Pi 5, but I haven't decided yet. Then I'll have a small server app that can pull the data from the Pi and draw some fancy charts.
The idea is to see trends and try to apply AI for correlating, at the first glance, completely unrelated data layers. Example how I'm thinking about this one: there's somewhat clear correlation that I sleep better when I do above average steps per day. How is my sleep quality affected if, let's say, I did above avg steps with a bad air quality at that time? (i.e. wild fires / pollen season / etc.)
I've built a Go application to ingest those data sources and currently finishing my first import use case - Apple Watch data.
Would be happy to connect and chat about this.
I talked more about it here: https://news.ycombinator.com/item?id=48881942
I see a lot of new (and, to be frank, a lot of mature ones) HR tools are just wrapping Chatgpt around resumes (almost like "OK, now match this resume against this job posting and tell me if applicant fits"), which introduces a massive bias/inference problem.
I decided to build the exact opposite – a deterministic, math-driven fitness engine. It extracts structured scorecards from both CVs and job requirements and mathematically matches them, so you can actually review the exact reasoning behind why a candidate scored a, say, 85%. This fitness value is specified at every interview step – as applicant goes through an interview process their scorecard is updated at all steps.
If anyone here builds in the HR space, I’d love your feedback.
One day I will make a game.
However, LLM coding has made coding less rewarding so… Im thinking about starting a new hobby as coding for fun has become prompting.
With all the supply chain attacks on OSS ecosystems targeting developers, PMG is a practical protection using a combination of threat intel, policy and sandbox.
It’s a package firewall on the terminal really. It has been surprisingly effective against most of the recent attacks.
A News Platform aggregator collecting sources of information across the internet (socials, newswires, etc.) and trying to push context to humans in a more digestible form. We are also experimenting with defining lineage of information using AI to help people try to piece the puzzle together as information flows in.
Have now the core done and working on a MVP UI to validate it.
One of the things I always wish to do properly was to model currency and unit of measure in full as core types, plus truly trace everything related to the business transaction from production to beyond the sale.
Looking into a persistent workflow engine like `temporal` now...
P.D: I'm debating if open source or not, in light of the AI-pocalypse...
It’s an AI-powered mock technical interviewing platform, for system design and coding.
I’m also working now on behavioral mocks, with a coach feature!
I’ve been working on it on and off for a year, but started spending significant time in the last few months.
I know everyone’s burnt out on LLM products, but I think it’s nice for this kind of prep since you can do it on demand and in an environment it’s safe to fail as much as you need without judgement so you can actually learn.
It’s early and free if anyone is interested in trying it out (at least while I can afford to serve it for free)
I went on a side quest to strip out ProseMirror and markdown-it and implement a custom stack instead. I open sourced both the parser and editor (https://saturn9.studio/technology/):
* Markoffset is a fast, plugin-based, incremental Markdown parser: https://github.com/saturn9studio/markoffset
* Scribeframe is a text editor engine: https://github.com/saturn9studio/scribeframe
It's opensource and more modern.
A text-based song format for generating music. I wanted to be able to create a song entirely using text, so I created a TOML-based format for doing so, and gave it most of the features you would find in a DAW. Since the format can be described in a SKILL file, AI can be used to write a song in this format, which can then be converted to audio.
It's called peek-cli: https://github.com/puffinsoft/peek-cli
Aggregator for new posts in build threads from 277 old-school DIY forums.
Build threads of people building cars, 4x4s, motorcycles, boats, airplanes, hot rods, musical instruments, etc.
Building a typing application that helps you quickly learn and improve your typing.
We believe everyone can type at 80wpm or more. It just takes a good tool and a couple months of consistent practice
Obviously this is going to take a bit more work but at least the resource usage will stay low, which I consider quite important. Especially since gamers are a large portion of the user base.
- lazyslurm: A TUI tool for managing/viewing slurm / HPC setups. Similar to lazygit or lazydocker (https://github.com/hill/lazyslurm)
It's for iPhone, and for the best experience, Apple Watch. It's very early, playable via TestFlight, and I would love feedback! There's a TestFlight link at: https://reverdure.yourstrategy.co
Fortunately I think I've been bailed out by agentic coding the last couple months from a product perspective but I think the major gains so far have been due to marketing and exploring alternative growth channels. Even so, keeping momentum is never a given and requires constant output from all angles! Onward...
Most recently, adding SID support, and adding timing information to the emulated formats that don’t have any tagged song duration (e.g., converting NSF to NSFE). This means playing the songs one by one and watching for repeated sequences of writes to the sound chip registers.
It's called TinyToT: https://github.com/guilt/TinyToT
You basically get a LLM without any training/RL here.
It's been fun dealing with memory and C's weird design in this age of agentic coding.
It’s basically snake meets scrabble meets PvP stealing. It’s a novel idea and I think it’s cool it hasn’t really been done before :)
The issue is it’s too complicated, the onboarding is dogwater, and the aesthetic is too complex
So I’ve spent the weekend fixing onboarding, fixing and relaxing the visuals mix and simplifying mechanics.
I’ve also tested LLMs playing the game through a harness I wrote. LLMs get smashed, they can form words and steal, but they lose badly to conventional bots.
I’ll be exposing an LLM leaderboard on my next release (hoping this weekend) with links to game replays for the LLMs.
Would love for people to give it a try, give me some feedback, and say what you’d love to see on the roadmap.
[0] - https://snibble.gg/
But everyone who has played has had the same feedback lol so that’s what I’ve been changing this weekend :)
Thanks heaps for trying it!
Completed games can be “replayed” and replays can fit in a QR code upto 30 minute games. So I think that’s pretty cool
- https://smacke.net/ffsubsync -- automagically synchronize subtitles, now purely client-side in your browser thanks to pyodide
- https://ipyflow.github.io/ipyflow/lab/index.html?path=demo.i... -- reactive python jupyter notebooks, again in the browser thanks to pyodide / jupyterlite
- https://smacke.net/pipescript/lab/index.html?path=demo.ipynb -- magritter-like pipe / placeholder syntax for ipython / jupyter, again able to run purely in the browser
- https://smacke.net/pycograd/lab/index.html?path=pycograd_sim... -- pyccolo and pipescript-powered autograd, once again able to run purely in the browser since numpy has a wasm target (notice a theme here :) )
So I started a new side project: decompilation of my cherished childhood video games. Many Mega Man games, starting with Mega Man Battle Network 2.
I just finished polishing and verifying the early initialization routines, and have already traced various parts of the game's engine. I was surprised to discover that it was a huge state machine of sorts. I want to focus on reverse engineering the saving system so I can write a save editor, and the music system so I can listen to the music.
Good luck with your build and perhaps I might get interested in future too as I did once have a thought that having a custom car to me would reflect more cool-ness than an expensive one. I am really interested by small cars, perhaps retro. I imagine my favourite car to be somewhat like the car that Ryan gosling drives in La La Land.
but a cool project nonetheless, certainly thinking about it inspires a bit of car enthusiasm within me even though I am not that much of a car fan so much right now so a really cool project if it can help more people feel this spirit. good luck :-D
I have a question but how does building new (retro-inspired?) cars go about in terms of pricing. I feel like they might be too costly to get custom-built and that If I really ever in my life go about doing this, I would prefer DIY but I still imagine that it might be too expensive or hard to make a car. Are there any go-to cars which are easy/recommended within this space and how does it compare off economically and what are the technical expertise that you require with this type of stuff?
Once again, I wish ya good luck in the project and would love to hear your answers for some of the questions I have!
You're right that getting a car custom-built is where the costs add up quickly; easily north of $50K. Most of the cost is labor, which is $0 if you do it yourself. Some of the projects are much easier than others. If you want to fall down a rabbit hole, look in the kit car and hot rod categories; lots of affordable and small builds in there. The Buick Riviera in La La Land is more of a resto-mod cruiser project, but the small/retro itch is exactly what the kit car category scratches. The first step is to find a forum where people are building the car you like, and start following related build threads. That's the majority of my social media intake these days.
We just launched a couple weeks ago and we’d love any feedback or suggestions!
Open to feedback and missing pieces.
i make the microsoft word but less sucks, and there is scientific calculator integrated and also ai on it too, available on linux (stable) and windous (unstable).
Realize that I'm really bad at marketing. Trying to work on it.
It lets you take a picture of video games and shows price comparisons for the major buy lists.
planning on postin ga show hn this week.
The core idea was that I've always been a lousy notetaker, even going back to my school days years ago. I'm great at one-off and one-liner notes and occasionally more in-depth notes, but tend to not flesh them out fully enough to make them worth re-visiting.
This has been a struggle even as an engineer sitting in meetings or trying to absorb new information when starting a new job and ramping up.
Logbook is meant to use an interaction paradigm we as engineers are using very often these days: it's a terminal UI in the vein of Claude Code, OpenAI Codex, etc.
It's targeted at the entry of free-flowing thoughts but you can also write longer notes by launching your default shell editor from within the tool.
Each note is saved as markdown with some metadata and that metadata is then saved to a local SQLite DB.
For the LLM side, the tool extracts useful metadata from those notes and then performs some local ranking/categorization. It then has the ability to send a note or some metadata to a provider of your choosing (it's straightforward to use OpenAI or something more broad and customizable like OpenRouter) for further enrichment or filtering.
A couple examples of the currently implemented slash-commands: `/related` can be used to find related notes; say you've been scribbling down notes about OAuth or MCP servers and want to gather up the most relevant notes to one of those topics. Or you can use a `/gaps` command that'll help you find things you've taken notes about but without properly defining or providing context around them (i.e. you mention ID-JAG for OAuth but never actually say what ID-JAG is, this command will tell you this so you have a chance to review what you previously wrote and can then define exactly what that keyword is about).
It's still very much a work in progress. It's not meant to be a full-fledged note-taking app a la Obsidian or anything like that. I've just always preferred taking notes in markdown or plain text and this is a great way to continue doing that while also making enrichment of the notes pretty simple.
You may ask "why not just use agent memories?" I don't really like the idea of tightly coupling notes with codebases or agents and I don't find the current UX very intuitive at least for the way I prefer to take notes.
A data layer to connect everything with everything
wafertown (few days old!)
World's first LLMORPG. You craft a prompt and it goes to live in wafertown and interact with other players (I mean prompts), you can change your prompt once per day, then next day you get news about what you did there!
Super early, everything is manual rn, I'm automating stuff including sign ups, if you want to join shoot me an email!
Search is currently provided by the Radio Browser API, but I'm now building my own station API with proper metadata and thumbnail coverage. A station discovery page with most played stations is also in the making.
Ask the same engine the same question twice and you get different answers, different citations, sometimes a different opinion of your brand, so figuring out how best to present this has been a fun product problem to solve.
It also tries not to be yet another dashboard: instead of just analytics, an agent turns the findings into a ranked list of "ship this fix" todo items.
I started them with ebitengine (Golang) but got somewhat frustrated with its web builds, and so built my own thing for small games that I want to work great on mobile or native PC, but also on web. I call it NanoGame, the host is written in Rust and the games are AssemblyScript. I've ported a number of other small games I had written to it as well, but haven't released any.
Two of the games I released a couple days ago were actually the ebitengine versions, but have partial ports to my framework, and the third I released the version using my stuff.
https://scramblequest.app - ebitengine, word search game where you slay monsters with the words, has a long campaign as well as a daily challenge and unlimited play
https://wordpeek.app - ebitengine, another word search game, this one reveals pieces of a picture and your goal is to guess the picture
https://playsilhouette.app - my own framework, this is a simple matching/hidden object(ish) game, more for kids
I also made a little umbrella site for them at https://playthese.wtf
Big bang prototypes have been pretty awful, even after feeding the LLMs huge documents / wishlists / descriptions of how it should work, etc. Part of the experiment was giving LLMs some leeway to make product decisions with a lot of north star guidance, but AFAICT they are really bad at this. I also tried basic bottom-up efforts, which have been better but obviously more tedious. Now I'm trying to find a more scalable bottom-up approach that is more LLM-accelerated
But maybe you should checkout the tools it’s based on, sem - https://github.com/Ataraxy-Labs/sem and ultimately treesitter. They at least give a more structured approach to dealing with code than simple text.
Earlier this year, a colleague encouraged me to experiment with Claude Code. So now I have a little game project. :) Being unfamiliar with genAI, I chose something modest so that I'd more likely be able to push it to a fairly polished state.
Tentatively called Vestiges, it's a single player 2D roguelite strategy game with meta progression, some narrative, and a card minigame (the latter inspired by work I did on Star Wars: Knights of the Old Republic II). It's set in the near future. You are using software (the game) to navigate a person's digitized mind, reading their memories.
I hope to have a playable demo within the next month or so.
2. My take on an agent framework ... append only log + content hypergraph in Elixir, tools that regularly pull data from other services into Postgres—built as a kind of 'exoskeleton' around claude/codex so it's not competing with fast-moving tools.
Thinking about category theoretic models of computation: https://arxiv.org/abs/2208.03817
--
Some things I want other people to build:
- Indexing for Github
- All-in-one social media ingestion libraries for agents
- GOFAI-inspired knowledge / semantic / research graph stuff—I want to point agents at rules/structures for writing connected, verifiable statements
Written by Codex with me driving product direction, reviewing, testing, occasionally scolding, and handling the release process.
Accepted onto the Mac App Store last week.
You can put your face on the screen in real time, record, stream, even annotate live, add text, draw, show touch indicators.
Pretty neat!
I am trying to involve family members' specialties and interests so she can elicit help from each person: entomology, mechanical engineering, etc.
All that for her to discover the Secret Planned Activity the following day (visiting a theme park.)
This doesn't work in simultaneous-move settings like Orbit Wars (or order-book markets), converging to an exploitable pure strategy rather than a Nash equilibrium.
LeCun's JEPA, by contrast, is a learned neural world model, which lacks the determinism, speed, and debuggability of a code-based simulator. Thus, it can drift or predict illegal states, and you can't inspect why it made a prediction the way you can trace a Python function.
TL;DR: The benefit is better auditability and easier RL-like training. The SM-MCTS extension fixes the first problem (decoupled UCB per player approximates Nash equilibrium instead of a pure strategy) while keeping the second advantage intact (a deterministic, inspectable code simulator).
https://github.com/ternary-ai/ow-code-world-model https://jdsemrau.substack.com/p/a-self-improving-code-world-...
New essays published every Wednesday.
Please subscribe if you enjoy it. It is and will stay free.
[1] https://www.amazon.com/dp/B0H7FLQDYD
[2] https://www.chestergrant.com/7-truths-from-veritas-by-cheste...
[3] https://www.chestergrant.com/what-different-cultures-agree-o...
Here is the repo where the work was happening: https://github.com/mnikic/hurd-journaling
The host made an offhand mention that there's probably a bunch of other similar sites that could be created with all the of useful but difficult-to-access government data out there. That sounded interesting, so I thought I'd give it a whirl!
Working on a few of them, including The Waterline (https://the-waterline.com/) for water info for the western US, The Scramble (https://the-scramble.com/) for egg prices, and The Dwell (https://the-dwell.com/) for container ship dwell times.
All pretty fascinating topics to learn about, plus it's been interesting to see how much of the website setup I can fully delegate to Claude. With Cloudflare to buy domains and put the sites up, a Google Service Account with access to Google Search Console and GA4 to create those properties and a Buttondown API key for weekly email sending, it's almost all hands off for me. Though it refuses to take control of the browser and create a new Buttondown account, which I was surprised is a red line.
https://azriel.im/disposition/
Things I missed in original graphviz dot:
1. predictable / stable layout
2. dark and light mode css (tailwind)
3. interactive through pure css
4. markdown descriptions
Took ages understanding how to route edges to not overlap labels.
I have been working on a set of tools and standard formulas that can be applied to these cases and demonstrate a more accurate view of a team's or department's overall ROI. The plan is to open-source the bulk of it, but provide a hosted service for folks who don't want to manage it themselves.
AI-first, MCP ready to host single HTML page. Connect & publish directly from ChatGPT app.